Predicting the past - Tilburg University, The Netherlands

namely a greenhouse effect of 1.87 0 C, a solar radiation effect of 1.09 0 C, and a remainder
term of 0.05 0 C. Hence, if aerosols and solar radiation had remained at the 1959 level,
then the expected global average temperature would have been 1.09 0 C higher. The solar
radiation effect is therefore important, masking 58% of the increase due to the greenhouse
effect. Ignoring dimming thus causes a serious underestimation of the greenhouse effect.
Our results are alarming. Fortunately, we are now in an economic crisis: less production,
less pollution, less global warming.
Tennis
Let us turn to a happier subject. It must have been in June 1992, during the period
that I was working both in London and in Tilburg, that Eveline and I were watching the
tennis at Wimbledon on TV and listening to the BBC commentators. We discussed and
subsequently kept a record of the statements made by the commentators, for example
that serving with new balls provides an advantage, that a ‘winning mood’ exists, or that
top players must grow into the tournament and that they perform particularly well at the
‘big’ points. We wondered whether these commonly accepted ideas were true or not, so I
contacted IBM to give me their data, which they did.
Two years later, in 1994, I was joined by Franc Klaassen, then a student here in Tilburg,
now full professor at the University of Amsterdam. For almost twenty years we did
research on tennis, first on these accepted ideas, later also on other — statistically more
demanding — aspects of tennis, such as the existence of winning mood, how efficient top
players are, and how we can best predict the outcome of a match, while the match is in
progress. In the end Franc and I wrote 15 papers together, most but not all about tennis.
Tennis research, like all sports research, may be considered a frivolous activity, and you
may well wonder why the taxpayer should support scientists engaged in such activities.
You are right, but only in part. Sports data have some characteristics that make them
ideally suited for studying human behavior. They are clean — there are few errors in the
data — and the data collection is transparent and can be checked. Data in economics,
psychology, and many other sciences are often messy and ambiguous. For economists,
psychologists and other scientists to work with clean data provides a great opportunity
and a welcome change from normal practice. If results do not come out the way they
‘should’, then this cannot be blamed on the data. There must be a ‘real’ explanation and
we have to find it. Maybe our preconceived idea is wrong or maybe we have not applied
the correct statistical method.
Predicting the past 21